Cerfacs Enter the world of high performance ...

The assessment of the impacts of climate change often requires to set up long chains of modeling, from the model to estimate the future concentration of greenhouse gases to the impact model. Throughout the modeling chain, sources of uncertainty accumulate making the exploitation of results for the development of adaptation strategies difficult. It is proposed here to assess the impacts of climate change on the hydrological cycle over France and the associated uncertainties. The contribution of each sources of uncertainty is not addressed, but mainly that associated with greenhouse gases emission scenario, climate models and internal variability.

In the context of impacts of climate change on the hydrological cycle over France, it is possible to ask what is the contribution of each sources of uncertainty to the total uncertainty associated with mean changes. Is it possible to reduce, and if so how, the contribution of one source or another ?

We propose in this work an approach to assess the transferability in the future climate of a statistical method to downscale climate simulations. The transferability assumption is one the main sources of uncertainty in statistical downscaling method. The assessment suggested here relies on the use of regional climate models, in a perfect model framework, and shows that some predictors are useful to ensure the transferability of the downscaling method in the future climate. This framework, proposed for a statistical downscaling method, is also applicable to bias correction methods in regional climate models.

Recent atmospheric reanalyses of the 20 h century are downscaled with the method developed in this work, associated with observations of temperature and precipitation. The hydrological cycle over France is characterized with these reconstructions. We show that the multi-decadal variability of observed stream ows during the 20th century is generalized to the whole country and is partly due to atmospheric variability. This multi-decadal variability of stream ows is generally weaker in hydrological simulations done with historical simulations from climate models.

The climate projections have been downscaled with the method developed in this work. The temperature on the country, on average over climate models, could increased by 3,5 C in winter and 6,5 C in summer in the course of this century. Precipitations will decrease all over the country in summer, nearly by half on southern part of France for the most severe scenario. In winter, precipitations will increase in the northern part of the country and will decrease slightly in the southern part. In the next few decades, the decrease in precipitation is important in summer, and changes are less pronounced for other seasons.

Results of hydrological projections done with one hydrological model and an ensemble of climate models are presented for the coming decades and for the end of the century. On the Seine river, results slightly differ in winter from those presented in previous studies. Here, precipitations and stream ow increase in winter and decrease in summer on that river basin. Elsewhere in France, results are consistent with previous studies, namely an increase in evapotranspiration, a decrease in stream ow and much drier soil.

The uncertainty due to both climate models and internal variability on relative changes in stream ows always increase during the 21 st century, to over 20% in winter for the most severe scenario. In the coming decades, the uncertainty due to internal variability only on stream ow changes is as strong as the uncertainty due to both climate models and internal variability. In the coming decades, annual stream ow changes of the Loire, Garonne and Rhône rivers are stronger than the maximum changes observed during the 20th century.